www.nature.

com/scientificreports

OPEN Abnormal synergistic gait

mitigation in acute stroke using
an innovative ankle–knee–hip
interlimb humanoid robot:
a preliminary randomized
controlled trial
Chanhee Park1,2, Mooyeon Oh‑Park3,4, Amy Bialek5, Kathleen Friel5, Dylan Edwards6,7 &
Joshua Sung H. You1,2*

Abnormal spasticity and associated synergistic patterns are the most common neuromuscular
impairments affecting ankle–knee–hip interlimb coordinated gait kinematics and kinetics in patients
with hemiparetic stroke. Although patients with hemiparetic stroke undergo various treatments to
improve gait and movement, it remains unknown how spasticity and associated synergistic patterns
change after robot-assisted and conventional treatment. We developed an innovative ankle–knee–
hip interlimb coordinated humanoid robot (ICT) to mitigate abnormal spasticity and synergistic
patterns. The objective of the preliminary clinical trial was to compare the effects of ICT combined
with conventional physical therapy (ICT-C) and conventional physical therapy and gait training
(CPT-G) on abnormal spasticity and synergistic gait patterns in 20 patients with acute hemiparesis.
We performed secondary analyses aimed at elucidating the biomechanical effects of Walkbot ICT
on kinematic (spatiotemporal parameters and angles) and kinetic (active force, resistive force, and
stiffness) gait parameters before and after ICT in the ICT-C group. The intervention for this group
comprised 60-min conventional physical therapy plus 30-min robot-assisted training, 7 days/week, for
2 weeks. Significant biomechanical effects in knee joint kinematics; hip, knee, and ankle active forces;
hip, knee, and ankle resistive forces; and hip, knee, and ankle stiffness were associated with ICT-C. Our
novel findings provide promising evidence for conventional therapy supplemented by robot-assisted
therapy for abnormal spasticity, synergistic, and altered biomechanical gait impairments in patients in
the acute post-stroke recovery phase.

Trial Registration: Clinical Trials.gov identifier NCT03554642 (14/01/2020).

The advanced research and development of innovative Robotic-Assisted Gait Training (RAGT) systems in the
field of robotic science have recently provided powerful and promising progress and, hence, hope for millions
of patients with synergistic hemiparetic gait after s­ troke1. Based on the contemporary task-oriented locomotor
learning theory, current stroke RAGT paradigms involve two commonly utilized systems (the Lokomat hip-knee
exoskeletal static RAGT, overground RAGT and G-EO end-effector RAGT) to mitigate the different aspects
of abnormal synergetic gait p ­ atterns1. The hip-knee exoskeletal static RAGT uses a top-down biomechanical
2
­model to focus on the hip and knee joint movements. The end-effector RAGT uses the bottom-up model,

1
Sports Movement Artificial‑Intelligence Robotics Technology (SMART) Institute, Department of Physical
Therapy, Yonsei University, 1 Yonseidae‑gil, Wonju, Gangwon‑do 26493, Republic of Korea. 2Department of
Physical Therapy, Yonsei University, Wonju, Republic of Korea. 3Burke Rehabilitation Hospital, White Plains, NY,
USA. 4Albert Einstein College of Medicine, Montefiore Health System, White Plains, NY, USA. 5Burke Neurological
Institute, White Plains, NY, USA. 6Moss Rehabilitation, Elkins Park, PA, USA. 7Edith Cowan University, Joondalup,
Australia. *email: neurorehab@yonsei.ac.kr

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 1

Vol.:(0123456789)
 www.nature.com/scientificreports/

emphasizing the ankle joint movement, which is often supported by a foot plate during locomotor ­retraining3–5.
The overground wearable RAGT (Ekso Ekso Bionics, Richmond, CA, USA) uses the bottom-up model, which
actuates movements of the hip and knee joints only for over-ground gait training in ­stroke6,7. While both exo-
skeletal, end-effector, and wearable RAGT systems have gained tangible improvements in gait function and
the associated biomechanical characteristics in patients with stroke 3,4,6,7, the important issue and underlying
synergetic gait problem remains unsolved and warrants further research and d ­ evelopment3,8. The synergetic
hemiparetic gait involves the loss of selective ankle–knee–hip joint movement coordination, which is associated
with abnormal spasticity, stiffness, and synergy due to cortical disinhibition post-stroke8–11. Clinically, the syn-
ergistic hemiparetic gait is classified as flexor and extensor and concurrently manifests with increased spasticity
and associated stiffness. The extensor synergetic gait is characterized by more increased ankle plantarflexion,
knee hyperextension, and hip internal rotation and extension along with a compensatory circumduction gait
when compared to normal c­ ontrols12–15. Specifically, the lack of open chain dorsiflexion in the terminal stance
results from dorsiflexor muscle weakness and spastic plantarflexors’ a­ ctivity13,16. Knee hyperextension in the
stance phase is observed to be compensating for the insufficient closed-chain dorsiflexion so that the tibia
rotates anteriorly, pivoting around the talocrural joint ­axis16. The quadriceps muscles are further weakened and
cannot support the knee and ankle during the stance ­phase17. Hip circumduction gait is a compensatory pattern
for iliopsoas and gluteus muscle weakness (50%) and the improper forward moment and longer level arm for
foot ­clearance16. On the other hand, the flexor synergetic gait is characterized by more increased external rota-
tion, abduction, and flexion of the hip (2.1°), flexion of the knee (19°), flexion (10°), and inversion of the ankle
than normal ­controls18,19. Insufficient plantarflexion occurs due to more gastrocnemius weakness and eccentric
motor control to advance the foot anteriorly during the terminal stance and early swing phases than normal
­controls12,13,18,19. The knee hyperflexion is associated with more quadriceps muscle weakness, hip hyperflexion
(6.5°), and external rotation (0.5°) due to the knee flexion during the swing phase when compared to normal
­controls14,20–23, ultimately leading to gait dysfunction in 85% of hemiparetic stroke ­population15. Therefore,
the present rationale for the robot-assisted training was to ‘break the abnormal ankle–knee–hip synergy’ or
improve selective ankle–knee–hip locomotor coordination in gait rehabilitation after stroke, rather than focus
on the amelioration of the hip-knee or ankle joint ­synergy3,24,25. In an extensive systematic review of the cur-
rent RAGT studies, patients with hemiparetic stroke were reported to exhibit an inherent abnormal synergistic
gait impairment, particularly in the ankle joint plantarflexor synergy even after intensive RAGT. However, the
overall gait function was e­ nhanced9,26. Such unresolved abnormal ankle synergy may have stemmed from the
insufficient locomotor coordination of ankle–knee–hip movement control in the currently used RAGT and
end-effector RAGT ­systems27,28. This scientific evidence corroborates the reported superior effects of RAGT
with ankle–knee–hip interlimb locomotor coordination control on volitional locomotor movement with “nor-
mal synergy” and motor control when compared to RAGT without ankle joint control (only knee-hip)24,29. As
such, stroke robotic rehabilitation clearly mandates for more effective and sustainable ankle–knee–hip interlimb
coordinated locomotor control to intervene on the synergistic gait impairment.
To overcome such shortcomings of the current exoskeletal (hip-knee control only) and end effector (ankle
control only) models, we have developed an innovative ankle–knee–hip interlimb coordinated humanoid robot
training (ICT) system (Walkbot, P&S Mechanics, Seoul, Republic of Korea). The ICT system is primarily designed
to create the optimal ankle–knee–hip interlimb coordinated locomotor movement, thereby mitigating such
underlying abnormal synergistic gait impairment in stroke ­rehabilitation26,30,31. This robotic system can detect the
patient’s current gait characteristics, such as the amount of participation or use in terms of active joint, angular
displacement excursion, active force/torque, and active weight-bearing center of pressure. The ICT system cal-
culates real-time ankle–knee–hip joint angles, joint moment, and muscle forces using a musculoskeletal anthro-
pometry model, including bone lengths, joints, inertial parameters tendon attachments. It can be personalized
to reflect subject-specific anatomic m ­ orphology26,30. Building on the contemporary motor learning theory of
task specificity, the ICT system allows accurate proprioceptive, kinematic, and kinetic guidance and real-time
motivational feedback concerning ankle–knee–hip kinematics and ­kinetics32. Importantly, ICT system enables
clinicians to provide variable error practice and high-intensity, repetitive, task-specific, and interactive exercises
of the paretic lower l­imb26,33. Recent ICT system empirical and clinical studies demonstrated excellent validity
(R2 = 0.86)34 and promising clinical improvements in balance and gait function (63.4%, 14.2%, and 15%) and
biomechanical characteristics (kinematics; 29.8%, 15.8% and 66.6%) in patients with hemiparetic stroke, spinal
cord injury, and cerebral palsy, ­respectively24,26,31,35–37.
Based on the conceptual framework of the ankle–knee–hip interlimb locomotor coordination on synergy
control, the present research has two specific aims. The primary purpose was to ascertain the therapeutic effects
of ankle–knee–hip Interlimb Coordinated robotic Training combined with Conventional physical therapy (ICT-
C; 30 min of ICT in addition to 60 min of physical therapy) on abnormal lower-extremity synergistic pattern,
which was determined using the standardized Fugl-Meyer Assessment of Lower Extremity (FMA-LE), when
compared to those of Conventional Physical Therapy and Gait training (CPT-G; 30 min of gait training + 60 min
of physical therapy) in patients with acute hemiparetic stroke. The secondary purpose aimed to investigate the
biomechanical changes associated with Walkbot ICT during acute rehabilitation, on kinematic (spatiotemporal
and angles) and kinetic (active force, resistive force, and stiffness) gait parameters, and to investigate the ICT-C
on spasticity which was determined using the Modified Ashworth Scale (MAS), compared to CPT-G in patients
with acute hemiparetic stroke. Correspondingly, our primary hypothesis was that there would be differences
in spasticity and abnormal lower-extremity synergistic pattern between the ICT-C and CPT-G. Our second-
ary hypothesis was that there would be significant differences in the kinematic and kinetic gait parameter data
between pre-and post-ICT intervention.

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 2

Vol:.(1234567890)
www.nature.com/scientificreports/

Characteristics CPT-G (n = 10) ICT-C (n = 10) P-value

Age (years) 70.60 ± 13.60 75.40 ± 11.21 0.749
Sex
Male (%) 3 (30) 5 (50)
0.206
Female (%) 7 (70) 5 (50)
Type of stroke
Onset time (days) 13.20 ± 7.20 7.60 ± 4.95 0.232
Ischemic (%) 10 (100) 10 (100) –
Side of hemiplegia
Left (%) 6 (60) 6 (60)
0.513
Right (%) 4 (40) 4 (40)

Table 1.  Demographic and clinical characteristics of the patients (N = 20). CPT-G conventional physical
therapy and gait training, ICT-C ankle–knee–hip interlimb coordinated humanoid robot combined with
conventional physical therapy.

Materials and methods

The present clinical research goals were twofold: The prirmairy goal was to determine the therapeutic effects of
ICT-C on abnormal lower-extremity synergistic pattern, which was determined using the standardized FMA-
LE, when compared to those of CPT-G in patients with acute hemiparetic stroke. The secondary goal was to
examine the biomechanical changes associated with Walkbot ICT during acute rehabilitation, on kinematic
(spatiotemporal and angles) and kinetic (active force, resistive force, and stiffness) gait parameters, and the ICT-C
on spasticity using the MAS, compared to CPT-G in patients with acute hemiparetic stroke.

Patients. A convenience sample of 20 patients with acute hemiparetic stroke (mean age 73.0 ± 12.72 years,
12 women) were enrolled as inpatients at the Burke rehabilitation hospital, New York, United states. The Albert
Einstein college of medicine institutional review board and the ethical committee (No. 2018-9283) approved
the experimental study protocol. After the patients were recruited via bulletin board notices within the hos-
pital, initial screening was conducted to determine whether the potential patients met the inclusion criteria.
Informed consent was obtained from all the patients before participation. This study was conducted by the rel-
evant guidelines/regulations and confirmed that informed consent was obtained from all patients and/or their
legal guardians. The study was conducted in accordance with the Declaration of Helsinki. The inclusion criteria
were as follows: (1) acute cortical/subcortical ischemic stroke (2 weeks post-stroke onset); (2) age between 18
and 99 years; (3) first clinical stroke presentation or prior stroke with no residual deficits affecting ambulation;
(4) ability to follow a two-step command; (5) Fugl-Meyer sensory score > 2; (6) suitability for gait training as
assessed clinically (ability to ambulate at least one step with a device/assistance); (7) height 132–200 cm; (8) hip-
knee joint length 33–48 cm; and (9) knee joint-to-foot length, 33–48 cm. The exclusion criteria were as follows:
(1) cerebellar/brainstem stroke, (2) body weight > 135 kg, (3) uncontrolled hypertension (stage 2) with blood
pressure > 160/100 mmHg; (4) cardiopulmonary impairments that can affect the ambulation test; (5) integumen-
tary impairment such as skin breakdown or bedsores around the suspension belt loading region; (6) relevant and
persistent mental illness; (7) lower-extremity fixed contracture or deformity; (8) bone instability (nonconsoli-
dated fractures, unstable spinal column, or severe osteoporosis necessitating treatment with bisphosphonates),
(9) other neurodegenerative disorders (amyotrophic lateral sclerosis, Parkinson’s disease); (10) MAS score > 3
in the affected leg; (11) relevant back or leg pain resulting in an inability to tolerate movement; (12) decreased
sensation impairing the ability to perceive whether the device is properly fitted, and (13) aphasia sufficient to
prevent the ability to communicate discomfort. Table 1 shows inter-group comparisons of baseline demograph-
ics and clinical characteristics of the patients. The nonparametric chi-square test for categorical variables showed
no significant differences in demographics or clinical characteristics between the groups.

Experimental procedure. A preliminary, randomized, single-blind, experimental design was used in the
present study. Coin flipping was used to assign patients to either the control or experimental group. A researcher
generated the random allocation sequence, another researcher assigned patients to interventions, and a third-
party blinded researcher assessed outcome measures. To remove experimental biases associated with the patients’
expectations, experimental information that could affect the patients was masked until the experiment was com-
pleted. A consistent experimental procedure was followed using intervention and standardized tests, including
MAS, and FMA-LE clinical measurements for both CPT-G and ICT-C groups before and after the intervention.
Additionally, biomechanical data including kinematic (e.g., angles), kinetic (e.g., active and resistive force), and
resistive stiffness in hip, knee, and ankle joints were measured before and after ICT-C. The same investigators
conducted all tests and interventions to improve the internal validity of the measurements.

Hardware. The hardware comprised an actuator module, a control module, and a power module. The actua-
tor module was rigidly attached to an exoskeleton and secured to the lower limbs using a velcro belt. The ICT
system was rigidly secured to the patients’ upper body (i.e., chest) using adjustable belts.

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 3

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 1.  The control scheme of the position-based impedance control for gait rehabilitation. ROB reaction
torque observer.

Actuator module. This module comprises two three-phase direct-current brushless motors, each provid-
ing output torque to the hip, knee, and ankle joints. The motors have a drive voltage of 24.0 V, a rated load cur-
rent of 2.0 A, and a maximum thrust load of 3.8 N.

Impedance control. The approach implemented for the ICT system was position-based impedance
c­ ontrol38. Mechanical impedance can be treated as the relationship between the force exerted by the actua-
tors and the resulting motion. As the mechanical impedance is viscoelastic, the restoring force is related to the
deviation of the robot’s reference trajectory and velocity. However, a dead-band and a limited threshold of angle
deviation were introduced to allow the normal variability of the human gait ­pattern39,40. The robot would only
intervene if the set level of trajectory deviation was exceeded.
The position-based impedance control law in joint space is given by

u = F̂(q)aq + ûext + Ĉ(q, q̇)q̇ + fˆ r(q̇) + ĝ(q) (1)

where ûext is the estimated external torque from the reaction torque observer.
The estimation of external torque is based on inverse dynamics
g    

ûext = ua + g F(q)q̇ + g F(q)q̇ + g F(q)q̇ (2)
s+g
g
where s+g is a lowpass filter and g is its cutoff frequency.
The acceleration term aq takes the following form:
aq = qd + K p e Imp + K ν ė Imp (3)

e Imp = qd − ûext Zd −1 − q (4)

ė Imp = qd − ûext sZd −1 − q (5)

where qd denotes the desired position and eImp, and ėImp denote the impedance error and its first derivative. Kv and
Kp ∈ ℜ2×2 denote the diagonal derivative and proportional controller gain matrices. Zd = Fds2 + Bds + Kd denotes
the desired impedance model.
Fd, Bds, and Kd ∈ ℜ2×2 are the desired mass matrix, damping matrix, and stiffness matrix. In (4) and (5), the
estimated torque feedback resulted in deviations of reference angular position and velocity. The overall scheme
of the force/torque sensorless position-based impedance control algorithm is shown in Fig. 138. The value of
mechanical impedance was manipulated by a therapist based on their experience, considering the patient’s move-
ment capability and disability levels. By adjusting the virtual mechanical impedance parameters, the therapist
could make the training more or less demanding for the patient. With a lower impedance, the patient needed to
participate more actively to maintain a functional gait pattern. In practice, only K was adjusted by the therapist,
and B would automatically adapt as a function of K.

Biomechanical measurements for kinematics, kinetics, and stiffness. The ICT system features a
biomechanical function to achieve efficient walking based on the inverted pendulum ­model38. Biomechanical
characteristics were determined using the kinematic and kinetic computing software (P&S Mechanics, Seoul,
Korea) of the ICT system, which calculates the angular displacement and active and resistive hip, knee, and ankle

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 4

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 2.  Lower-extremity kinematic joint angle calculation in ICT system. ICT innovative ankle–knee–hip
interlimb coordinated humanoid robot.

joint forces and t­orques38. Kinematic and kinetic data were synchronously obtained from each of the five gait
cycles in a steady-state, lasting > 5 min, from all patients before and after the intervention.
Kinematic measurements encompassed the joint angle, angular velocity, and acceleration, which were then
used to calculate the moment or torque associated with the body segment’s active and resistive forces acting on
the ankle, knee, and hip joints of the participant during walking. For example, when in full extension, the hip is
defined as at 0° flexion. When the thigh moves in an anterior direction relative to the pelvis, the hip is defined
as being in flexion (Fig. 2)41.
Hip angle = θh = θ10 − θ21

If θ10 > θ21, the hip is in flexion; if θ10 < θ21, the hip is extended.
Knee angle = θk = θ21 − θ43

Ankle angle = θa = θ43 − θ65 + 90◦

The convention for the ankle was slightly different, in that 90° between the leg and the foot was the boundary
between plantarflexion and dorsiflexion. If θa is positive, the foot is in plantar flexion; if θa is negative, it is in
dorsiflexion. Kinematic data were collected using a built-in potentiometer in the Walkbot system with a sample
rate of 36 Hz.
Kinetic measurements included active and resistive forces and torques of the body segment acting on the hip
joint during robotic interactive gait training. With the thigh lever arm acting on the robot system, the recorded
force data can be converted into hip joint torques acting between the ICT system and the participant’s leg. The
ankle–knee–hip joint torque data were collected by the servomotors mounted in the robotic system, in which
the corresponding encoders modulated the hip, knee, and ankle joint ­kinetics38. Specifically, the active force
was defined as a positive directional rotation force occurring in line with the target movement direction. In
contrast, the resistive force was described as a negative directional rotation force acting against the target move-
ment ­direction38.
g
The force equation is expressed as ûext = s+g (ua + g F  (q) q̇) − g F
 (q)q̇ + F  (q)q̇.
Clinically, an increase in force on the affected side represented an increase in voluntary strength recovery
of the paretic lower extremity. In contrast, a high resistance force indicated opposition that constrained active
limb movement during gait.
Furthermore, the kinematic and kinetic computing software of the ICT system was used to examine the
ankle, knee, and hip joint stiffness associated with RAGT. Graphical data were analyzed using a maximal sam-
pling rate of 72 Hz (gait cycle varies with the customized preferred walking velocity; frequency range 28–72 Hz
at 1.00–2.60 km/h) using a moving averaging filter. The stiffness k in the hip-knee joint-segment indicated the
resistance provided by an elastic body segment to deformation. Spasticity-related stiffness was computed based
on the joint angular displacement and resistive torque data, using a linear regression equation during the gait

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 5

Vol.:(0123456789)
 www.nature.com/scientificreports/

cycle 31, which was expressed as k stiffness = θF , where F is the resistive force acting on the knee, hip, and ankle joints;
and θ is the angular displacement produced by the force acting on the corresponding joint. In essence, a lower
stiffness value (approximately “0 or negative value”) represented a more significant active movement.

Clinical spasticity assessment. The MAS is a commonly used quantitative measure of spasticity or
muscle tone in response to passive limb m ­ ovements42. The ankle, knee, and hip flexors and extensors of the
paretic limb were tested according to a standardized ­procedure43. The grading ranged from 0 (“normal tone”)
to 4 (“rigid”). The MAS has been reported to be significantly responsive in detecting changes in muscle tone in
patients with hemiparetic stroke, and its reliability (weighted kappa = 0.87, standard error = 0.03, P < 0.001) has
been well ­established43.
The FMA-LE synergy scale (sub-score II index) was used to examine the lower-extremity sensorimotor
function and ankle–knee–hip joint function because it represents volitional or voluntary locomotor movement
patterns, which include flexor and extensor synergy. The flexor synergistic movement pattern comprised maxi-
mal hip flexion (abduction/external rotation), maximal knee flexion, and maximal ankle flexion. In contrast,
the extensor synergistic pattern consisted of hip extension/adduction, knee extension, and ankle plantarflexion.
Resistance was applied to ensure active movement and to evaluate both movement and strength. The ordinal
grading scale consisted of values as follows: 0 “cannot perform,” 1 “can partially perform,” and 2 “can completely
perform.” Clinically, 0 and 1 indicate an abnormal movement synergistic pattern, whereas a score of 2 indicates
the normal volitional movement synergy constituting the normal locomotor pattern. The total sub-score ranged
from 0 to 6 points for the volitional movement with the flexor synergy test and 0 to 8 points for the volitional
movement with the extensor synergy t­ est44. The reliability and validity of the kinematic and kinetic measurements
in ICT system were well established, intraclass correlation ­coefficient3,k = 0.96, and r = 0.65–0.93, ­respectively34,38.

Intervention. Both groups underwent an additional session of 30 min of therapy daily, 7 days/week, for
2 weeks. The CPT-G group underwent general inpatient treatment, including at least one 60-min session of
physical therapy per day. An additional 30-min standard physical therapy session was executed in the pre-ambu-
latory phase and/or for gait training activities. CPT-G was based on neurodevelopmental approaches and was
conducted by skilled and experienced physical therapists. The ICT-C group underwent general physical therapy,
which included at least one 60-min physical therapy session and the additional 30-min ICT session. Anthro-
pometric data, including height, weight, foot size, thigh length, shank length, and ankle height, were measured
and entered into the participant database. These data were used to automatically adjust the length and optimal
gait cycle of the exoskeleton legs according to each participant’s conditions. This provided the patients a sense
of safety using the suspension vest secured with elastic straps and connected to the harness mounted on the
counterweight system. Depending on the initial clinical conditions of the participant (e.g., pain, muscle weak-
ness, spasticity, tolerance, fatigue, or endurance), approximately 40%–60% (adjustable range, 0%–100%) of the
total body weight was sustained in the first session, which was gradually reduced by 5%–10% over the sessions.
The guidance force mode in the ICT system was used to increase the active engagement during robot-assisted
gait retraining accurately. According to the participant’s ability to improve beyond the initial target level (e.g., 40
Nm), the ICT system interactively adjusted the walking speed and resistive torque parameters based on patient
comfort and preference while attempting to minimize kinematic trajectory errors. The assistance guidance force
was systematically reduced from 100% (passive mode) to 0% (active mode). In the active mode, the system could
compensate for the weight, resistance, and inertia of the hemiparetic leg to achieve symmetrical, optimal gait
patterns. Furthermore, it provided real-time feedback from the treadmill, such as gait kinematics (joint angles),
kinetic forces (active, resistive torque, and stiffness) on the ankle–knee–hip interlimb coordinated movement,
and active torque on the ankle joint movement. During each session, the patients were provided with constant
verbal encouragement based on the results of real-time kinematic and kinetic data to optimize their gait pat-
terns. The ICT system was provided with virtual reality (VR)/augmented reality (AR) games (e.g., a virtual side
scrolling game Jordan jumping and taking the coins) and AR scenes (e.g., three-dimensional walking to explore
a king’s castle) to maximize the patient’s interest, motivation, and active engagement, while decreasing anxiety
and depression during the ICT session (Fig. 3)26.

Statistical analyses. Statistical data were expressed as means (M) and standard deviations (SD). The pre-
sent preliminary clinical study involved a non-superiority design in which the two-way analysis of variance
(ANOVA) and paired t-test were performed separately. The two-way analysis of variance (ANOVA) was applied
for MAS and FMA-LE data. Significant differences between the control and experimental groups were subjected
to Tukey’s post-hoc test. The paired t-test was used to compare the biomechanical characteristics (kinetics, kine-
matics, and stiffness) between pre-ICT and post-ICT in the experimental group. The Chi-square test was used to
analyze categorical demographic variables. Continuous variables were analyzed using the Kolmogorov–Smirnov
test. Independent t-tests were used to compare general characteristics of the patients between the groups. Addi-
tionally, Spearman’s rank correlation was used to determine the correlation among the MAS score, FMA-LE,
and stiffness. Based on our previous study, a power analysis using G-Power software (G-power software 3.1.9.4;
Franz Faul, University of Kiel, Germany) was performed to compute the minimum sample size r­ equirement31.
The sample size was determined to be 30 based on the effect size (eta squared, η2 = 0.6) and power (1 − β = 0.80)
on minimal clinically important difference (MCID) of FMA-LE and from torque and force d ­ ata31. SPSS for Win-
dows (version 25.0; IBM Corp., Armonk, NY, USA) was used, with a significance level set at α = 0.05.

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 6

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 3.  Flow chart. CPT-G conventional physical therapy and gait training, ICT-C ankle–knee–hip interlimb
coordinated humanoid robot combined with conventional physical therapy.

Results
Kinematic data. The paired t-tests showed that the mean post-ICT knee joint angle (M = 26.69, SD = 1.10)
was more increased than the mean pre-ICT knee joint angle (M = 22.42, SD = 0.61; t (9) = − 14.59; P = 0.00) in the
ICT-C group, indicating improved knee joint movement after ICT-C in patients with hemiparetic stroke (Fig. 4).

Kinetic data. The paired t-tests revealed that the mean post-ICT hip active force (M = 1.32, SD = 0.52; t
(9) = − 2.56; P = 0.03) was significantly greater than the mean pre-ICT hip active force (M = 0.59, SD = 0.48) in
the ICT-C group. The paired t-tests revealed that the mean post-ICT knee active force (M = 1.66, SD = 1.95; t
(9) = − 2.47; P = 0.04) was significantly greater than the mean pre-ICT knee active force (M = 0.05, SD = 0.04) in
the ICT-C group. The paired t-tests revealed that the mean post-ICT ankle active force (M = 1.52, SD = 1.06; t
(9) = − 2.71; P = 0.02) was more increased than the mean pre-ICT ankle active force (M = 0.46, SD = 0.67) in the
ICT-C group, indicating an improved hip–knee–ankle joint coordinated force after ICT-C. The standardized
effect size index, d, ranged from 0.64 to 0.67, indicating large clinical effects (Table 2).
The paired t-tests showed that the mean post-ICT hip resistive force (M = 2.08, SD = 0.11; t (9) = 61.61; P = 0.00)
was significantly greater than the mean pre-ICT hip resistive force (M = 6.18, SD = 0.21) in the ICT-C group. The
paired t-tests indicated that the mean post-ICT knee resistive force (M = 0.12, SD = 0.09; t (9) = 5.19; P = 0.001)
was more increased than the mean pre-ICT knee resistive force (M = 1.53, SD = 0.80) in the ICT-C group. The
paired t-tests revealed that the mean post-ICT ankle resistive force (M = − 0.07, SD = 0.53; t (9) = − 4.80; P = 0.001)
was significantly greater than the mean pre-ICT ankle resistive force (M = − 0.84, SD = 0.21) in the ICT-C group,
indicating an improved hip–knee–ankle joint coordinated force after ICT-C. The standardized effect size index,
d, ranged from 0.85 to 1.00, representing large clinical effects (Table 2).
The paired t-tests showed that the mean post-ICT hip stiffness (M = 0.72, SD = 0.17; t (9) = 1.32; P = 0.00) was
significantly greater than the mean pre-ICT hip stiffness (M = 1.53, SD = 0.23) in the ICT-C group. The paired
t-tests revealed that the mean post-ICT (M = 0.70, SD = 0.15; t (9) = 7.31; P = 0.00) was more increased than
the mean pre-ICT knee stiffness (M = 1.17, SD = 0.11) in the ICT-C group. The paired t-tests revealed that the
mean post-ICT ankle stiffness (M = 0.40, SD = 0.11; t (9) = 2.34; P = 0.04) was significantly greater than the mean
pre-ICT ankle stiffness (M = 0.67, SD = 0.33) in the ICT-C group, indicating an improved ankle, knee, and hip
joint coordinated force after ICT-C. Moreover, the standardized effect size index, d, ranged from 0.68 to 0.95,
suggesting large clinical effects (Table 3).

Clinical spasticity and FMA‑LE synergy data. ANOVA showed significant differences in the hip exten-
sor and ankle dorsiflexor MAS scores between the CPT-G and ICT-C groups (P = 0.000; 0.043). The post-hoc

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 7

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 4.  Paretic hip and knee angle kinematics in ICT-C group (unit: degree). ICT-C ankle–knee–hip
interlimb coordinated humanoid robot combined with conventional physical therapy; *Denotes significance at
P < 0.05; Number, mean; Bar, standard deviation.

ICT-C
Pre-test Post-test Mean difference t-value P-value Effect size
Hip active force 0.59 ± 0.48 1.32 ± 0.52 0.73 − 2.56 0.03* 0.64
Knee active force 0.05 ± 0.04 1.66 ± 1.95 1.61 − 2.47 0.04* 0.64
Ankle active force 0.46 ± 0.67 1.52 ± 1.06 1.06 − 2.71 0.02* 0.67
Hip resistive force 6.18 ± 0.21 2.08 ± 0.11 − 4.1 61.61 0.02* 1.00
Knee resistive force 1.53 ± 0.80 0.12 ± 0.09 − 1.41 5.19 0.001* 0.87
Ankle resistive force − 0.84 ± 0.21 − 0.07 ± 0.53 − 0.77 − 4.80 0.02* 0.85

Table 2.  Comparison of active force data in the paretic limb in ICT-C (unit: N). ICT-C ankle–knee–hip
interlimb coordinated humanoid robot combined with conventional physical therapy; *Denotes significance at
P < 0.05.

ICT-C
Pre-test Post-test Mean difference t-value P-value Effect size
Hip stiffness 1.53 ± 0.23 0.72 ± 0.17 − 0.81 9.16 0.00* 0.95
Knee stiffness 1.17 ± 0.11 0.70 ± 0.15 − 0.47 7.31 0.00* 0.87
Ankle stiffness 0.67 ± 0.33 0.40 ± 0.11 − 0.27 2.34 0.04* 0.68

Table 3.  Peak passive stiffness between pre- and post-test in a paretic hip, knee, and ankle stiffness (unit: Nm).
ICT-C ankle–knee–hip interlimb coordinated humanoid robot combined with conventional physical therapy;
*Denotes significance level at P < 0.05.

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 8

Vol:.(1234567890)
www.nature.com/scientificreports/

CPT-G ICT-C P-value

Pre-test Post-test Pre-test Post-test Time effect Between groups Time × group Effect size
Hip flexor 0 0.14 ± 0.38 0.31 ± 0.59 0.13 ± 0.35 0.837 0.077 0.107 0.003
Hip extensor 0 0 0.44 ± 0.62 0.25 ± 0.46 0.335 0.000* 0.335 0.011
Knee flexor 0.14 ± 0.38 0.36 ± 0.63 0.50 ± 0.53 0.25 ± 0.46 0.698 0.555 0.368 0.004
Knee extensor 0.14 ± 0.38 0.29 ± 0.49 0.31 ± 0.60 0.25 ± 0.46 0.678 0.580 0.335 0.005
Ankle dorsi-
0.29 ± 0.76 0.31 ± 0.59 0.14 ± 0.38 0 0.635 0.043* 0.527 0.006
flexor
Ankle plantar-
0.21 ± 0.57 0.14 ± 0.38 0.37 ± 0.74 0.13 ± 0.35 0.187 0.565 0.466 0.009
flexor
Flexor synergy 4.50 ± 1.20 5.50 ± 0.53 3.86 ± 1.83 5.29 ± 0.99 0.000* 0.116 0.615
Extensor
6.50 ± 1.85 7.50 ± 1.07 5.71 ± 2.16 7.07 ± 1.00 0.007* 0.170 0.797
synergy
Total synergy 10.00 ± 1.41 14.00 ± 2.56 9.14 ± 2.25 12.79 ± 2.64 0.000* 0.057 0.513

Table 4.  Modified Ashworth scale and Fugle-Meyer assessment lower extremity. MAS modified Ashworth
scale, CPT-G conventional physical therapy and gait training, ICT-C ankle–knee–hip interlimb coordinated
humanoid robot combined with conventional physical therapy, FMA-LE Fugl-Meyer assessment lower
extremity; *Denotes significance level at P < 0.05.

MAS spasticity
Stiffness Hip flexor Hip extensor Knee flexor Knee extensor Ankle dorsiflexor Ankle plantarflexor
Hip 0.459 0575* 0.279* 0.169 0.234 0.041
Knee 0.204 0.124 0.592* 0.697* 0.271 − 0.039
Ankle − 0.037 − 0.014 0.025 0.168 0.684* 0.600*
FMA synergy
Flexor synergy Extensor synergy
Hip − 0.01 0.01
Knee 0.16 − 0.18
Ankle − 0.42* − 0.31
MAS spasticity
FMA synergy Hip flexor Hip extensor Knee flexor Knee extensor Ankle dorsiflexor Ankle plantarflexor
Flexor synergy − 0.14 − 0.04 − 0.43* 0.01 − 0.41 − 0.14
Extensor synergy 0.01 − 0.22 − 0.01 − 0.43* − 0.27 − 0.07

Table 5.  Spearman’s rank correlation between MAS, stiffness, and FMA synergy. MAS modified Ashworth
scale, FMA Fugl-Meyer assessment; *Aenotes within-group significance at P < 0.05.

analysis confirmed more decreased hip extensor and ankle dorsiflexor spasticity after ICT-C than CPT-G, sug-
gesting that patients with hemiparetic stroke had decreased muscle spasticity after ICT-C but not after CPT-G
(Table 4).
ANOVA failed to yield a significant difference in the FMA-LE synergy scale score between CPT-G and ICT-C
(P = 0.12, 0.17; Table 4).

Correlation among MAS, FMA‑LE, and stiffness. A strong correlation was observed between knee
stiffness and knee extensor spasticity during gait (r = 0.70, P = 0.03) in the ICT-C. The correlation between ankle
dorsiflexor spasticity and ankle stiffness (r = 0.68, P = 0.02) and ankle plantar flexor spasticity and ankle stiffness
(r = 0.60, P = 0.04) were moderate. A moderate correlation was also found between hip extensor spasticity and
hip stiffness (r = 0.28, P = 0.04; Table 5). Additionally, the correlation between hip stiffness and knee flexor spas-
ticity was moderate. A moderate negative correlation was observed between flexor synergy and ankle stiffness
(r = − 0.43, P = 0.04). The correlation between knee flexor spasticity and flexor synergy and that between knee
extensor and extensor synergy was moderate (r = − 0.43, P = 0.02; Table 5).

Discussion
To the best of our knowledge, this is the first randomized controlled trial (RCT) on Walkbot-based RAGT to
evaluate comparative effects of ICT-C and CPT-G on clinical spasticity and abnormal synergy control in patients
with acute hemiparetic stroke and biomechanical effects of Walkbot ICT on kinematic (spatiotemporal param-
eters and angles) and kinetic (active force, resistive force, and stiffness) gait parameters before and after ICT.
As hypothesized, ICT-C demonstrated more positive effects in clinical spasticity and abnormal synergy control
than CPT-G alone. ICT was associated with positive effects on both kinematic (spatiotemporal parameters

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 9

Vol.:(0123456789)
 www.nature.com/scientificreports/

and angles) and kinetic (active force, resistive force, and stiffness) gait parameters. Most importantly, the pre-
sent robotic interactive gait training provided optimal ankle–knee–hip inter-joint coordinated training, which
reduced spasticity and associated stiffness and abnormal synergistic (extensor) gait patterns while improving
the active participation and associated active force during gait.
The kinematic analysis demonstrated that mean hip and knee joint angular displacements had improved by
8% and 16%, respectively, as a function of ICT-C in our cohort with acute hemiparetic stroke. Before RAGT,
increased compensatory hip hiking and circumduction gait were observed during the swing phase (owing to lim-
ited hip triple ankle–knee–hip flexion along with forward momentum). An 11% decrease in ankle dorsiflexion at
initial contact, hip extension, and limited ankle plantarflexion was evident at the terminal stance. After the robotic
intervention, the lower-extremity extensor synergy pattern accompanying circumduction and hip hiking was
substantially diminished because increased hip triple ankle–knee–hip flexion along with forward momentum was
observed during the swing phase. In contrast, an 8% increase in ankle dorsiflexion at initial contact, hip exten-
sion, and ankle plantarflexion at the terminal stance was remarkably achieved. Conversely, those who exhibited a
lower-extremity flexor synergy pattern accompanying gluteus maximus weakness and excessive hip circumduc-
tion showed a substantial reduction in this pattern because of the triple interlimb coordination of ankle–knee–hip
extension along with the extensor momentum were observed during mid-stance. With a 3% increase in ankle
plantarflexion during the pre-swing phase, hip flexion and ankle dorsiflexion during the loading response phase
were achieved. Our findings were consistent with previous kinematic studies that reported 3%, 10%, and 15%
improvements in hip, knee, and ankle joint angular displacements, respectively, after RAGT in 21 patients
with chronic hemiparetic ­stroke29. Bonnyaud et al. reported that RAGT using Lokomat improved the paretic
hip (1.3°), knee flexion (1.9°), and ankle plantarflexion (0.6°) range movement compared with conventional
gait training after 20 min in 26 patients with hemiparetic ­stroke45. Interestingly, ankle dorsiflexion remained
unchanged. Such a lack of ankle dorsiflexion might have resulted from the fact that Lokomat does not have an
independent ankle actuator that can help accurately modulate the ankle dorsiflexion-plantarflexion movement
in coordination with hip and knee joint movements during ambulation. A significant difference between the
Lokomat and Walkbot exoskeletal systems is an independent ankle actuator in Walkbot. The Lokomat system
is equipped with hip-knee joint actuators, which provide RAGT focusing on hip and knee joint movements. In
contrast, the ICT system comprises ankle–knee–hip joint actuators, which allow the natural ankle–knee–hip
interlimb coordinated movement during ­locomotion9,26. Therefore, the underlying rationale for such kinematic
improvement may be associated with the fact that RAGT provided corrected gait-specific kinematic (kinesthetic)
feedback to the spastic muscles that were synergistically activated into hip and knee extension during the stance
phase of walking and hampered the flexion of the hip and knee for foot c­ learance29,46. Little et al. observed that
abnormal hip hiking and circumduction affected foot clearance owing to limited hip and knee flexion kinematics
compared with limited ankle dorsiflexion kinematics alone. This highlights the importance of ankle–knee–hip
inter-joint coordinated locomotor c­ ontrol47.
The kinetic analysis revealed substantial enhancements in active force, resistive force, and stiffness in the hip
(55%; − 197%; − 113%), knee (97%; − 92%; − 67%), and ankle (70%; − 13%; − 68%) putatively as a function of
ICT-C. Notably, gait-related active force gains in the hip, knee, and ankle joints improved by 13%–197% in ICT-C.
This finding substantiates previous RAGT evidence using Walkbot, demonstrating more substantial improve-
ments in hip flexion (1.05 Nm) and extension (0.16 Nm) active torques, hip flexion (− 0.56 Nm), and extension
(− 0.26 Nm) resistive torques, and hip flexion (− 0.54 Nm/°) and extension (−0.2 Nm/°) resistive stiffness as a
function of the intervention in patients with subacute ­stroke31. Notably, such positive changes induced by RAGT
in active torque, resistive torque, and associated stiffness ranged from 20 to 80% in patients with chronic stroke
who exhibited inherently increased joint stiffness and muscle shortness associated with spasticity despite pro-
longed deconditioning after stroke o ­ nset31. Certainly, it is plausible that such paradoxical phenomena between
active and resistive forces or stiffness support the commonly held premise of reciprocal inhibition of spastic
plantarflexors and associated extensor synergetic gait pattern impairment after ­stroke48,49. The kinematic and
kinetic findings reported in the literature support the notion that RAGT using just hip-knee joint, or ankle–foot
plate actuators alone cannot mitigate the extensor or flexor synergetic gait impairment in stroke rehabilitation.
Close biomechanical coordinated coupling between the ankle, knee, and hip joints is recognized in a healthy
gait ­pattern41, whereas exaggerated hip flexion, hiking, or circumduction synergy is commonly used to clear the
toe as a compensatory manifestation in a stereotypical hemiparetic gait ­pattern10. Given the dynamic role of the
ankle–knee–hip interlimb locomotor coordination, our findings are consistent with those of other studies sug-
gesting that ankle dorsiflexion and plantarflexion torques help in limb ­advancement10,50,51. Recent ankle robotic
biomechanical evidence confirmed that ankle robotic assistance helped generating sufficient ankle dorsiflexion
and plantarflexion kinematic (7°) and torque values (20 Nm), which play a cardinal role in the paretic limb
forward advancement moment (22 Nm) of patients with hemiparetic synergetic gait i­ mpairment3,10,51,52. In the
present robotic paradigm, the ICT detected altered biomechanical characteristics associated with spasticity and
synergistic gait patterns that were initially guided and facilitated based on the real-time kinematic and kinetic
feedback about the ankle–knee–hip locomotor movement. Such locomotor movement sense feedback is essential
for proprioceptive sensory awareness required during the locomotor skill relearning because the majority of the
patients with hemiparetic stroke experience altered sensorimotor function following the sensorimotor cortex
lesion. The accurate sensory inputs about locomotor related joint angle and force are transmitted to the spinal
cord, subcortical level, and cortical level of the sensorimotor cortex via the ascending proprioceptive pathways
(dorsal column and medial lemniscus)53,54. The locomotor related sensory feedback involves the modulation of
the supraspinal networks including subthalamic locomotor region, cerebellar locomotor region, and descending
pontine locomotor regions, and mesencephalic locomotor regions, where locomotor signals are transmitted via
pontine reticulospinal pathway to the spinal cord central pattern generators, thereby regulating the interlimb
coordinated locomotor p ­ attern55,56.

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 10

Vol:.(1234567890)
www.nature.com/scientificreports/

Additionally, task-oriented locomotor re-learning was progressively challenged by increasing the amount
of active participation or the use of the paretic and nonparetic limbs (1000 repetitions or steps), which were
underutilized in the conventional therapeutic approach (292.5 steps)57. Furthermore, our clinical FMA-LE and
MAS data revealed that the abnormal synergy pattern was significantly reduced along with improvements in
spasticity in the ankle and hip joints, as evidenced by increased volitional movement with synergy in FMA-LE.
Correlational statistics demonstrated that the hip, knee, and ankle joints’ resistive stiffness was moderately
related to spasticity in the hip extension, knee extension, ankle dorsiflexion, and ankle plantarflexion muscles
and inversely correlated with volitional movement with synergy in FMA-LE (Table 4). These results corroborate
the classical relationship between spasticity and stiffness, as well as the synergistic ­pattern58. Ankle spasticity
has been consistently reported to be associated with ankle stiffness (r = 0.23) and abnormal synergy in adults
with spastic hemiparetic g­ ait59. To date, no clinical evidence exists regarding the unpinning neuromechanical
relationships between spasticity and stiffness, as well as the synergistic pattern during locomotion when RAGT
is implemented. It is plausible that enjoyable (VR), active, repetitive locomotor movement (1000 repetitions or
steps) via ICT can facilitate agonistic activation (dorsiflexion) while reciprocally inhibiting abnormal spasticity
and synergistic antagonist activation (e.g., plantarflexion) during the initial contact of the gait c­ ycle57. Abnormal
spasticity and associated stiffness in the ankle plantar flexors are strongly influenced by stretch reflex hyperex-
citability as a result of cortical disinhibition in adults with hemiparetic ­stroke59, which generates a stereotypical
asymmetrical extensor synergistic gait p ­ attern46. Certainly, such an altered synergistic gait pattern was mitigated
by active, enjoyable, repetitive locomotor movement via ICT combined conventional physical therapy, which
may be related with neuroplasticity and associated locomotor functional r­ ecovery3,8,26,30. In fact, our patients’
post-intervention survey reported that the ICT was enjoyable and motivating. The technology acceptance was
previously evaluated by conducting the Participant Satisfaction Questionnaire, which yielded that most patients
who successfully completed the RAGT reported that the Walkbot RAGT was safe, fun, and beneficial for gait
­training26. Similarly, our previous functional magnetic resonance neuroimaging study validated that locomotor
training using VR increased blood oxygenation level-dependent signals (0.6) in the ipsilesional primary sen-
sorimotor cortex in patients with chronic hemiparetic ­stroke60. Limitations of the current research should be
considered in future investigations. A major limitation is that although the present results are promising, they
should be interpreted carefully when attempting to extrapolate the current findings to clinical practice and to the
management of stroke participant rehabilitation due to the small sample size. The initial sample size was proposed
to be 30 patients while accounting for the 25% attrition rate. However, due to the complicated nature of patients
with acute stroke, only 20 of 30 (66%) patients who successfully completed the pre-test, intervention, and post-
test were included in the final data analysis. The remaining patients could not complete the test due to fatigue,
other comorbidities, medical complications, or early discharge. Another limitation is that biomechanical data
were only obtained from the ICT-C group because biomechanical assessment outcome data may be influenced
by the likelihood of the ICT-C group to be more familiar with the Walkbot biomechanical assessment than the
CPT-G group. Nevertheless, in future research, it would be more appropriate to implement the biomechanical
assessment for both groups after sufficient familiarization of the Walkbot biomechanical assessment.
In conclusion, ICT involving ankle–knee–hip movements together with conventional physical therapy in the
acute inpatient phase after stroke was associated with improved biomechanical gait profile and clinical status.
Moreover, our correlational statistics indicated that the hip, knee, and ankle joints’ resistive stiffness moderately
correlated with spasticity in the hip extension, knee extension, ankle dorsiflexion, and ankle plantarflexion mus-
cles and inversely correlated with volitional movement with synergy in FMA-LE in the ICT-C group. Our results
highlight the incorporation of ICT with conventional therapy as a successful intervention for abnormal spastic-
ity, synergistic, and altered biomechanical gait impairments in patients with acute stroke. Most importantly, the
Walkbot ICT system allows an autonomous liberty to provide accurate real-time quantitative biomechanical
feedback as well as an effective and sustainable ankle–knee–hip interlimb coordinated locomotor training, which
could serve as a basis for advanced robotic science and medical research.

Received: 14 April 2021; Accepted: 3 November 2021

References
1. Kim, H. Y., Park, G. L., Shin, J. H. & You, S. H. Neuroplastic effects of end-effector robotic gait training for hemiparetic stroke: A
randomised controlled trial. Sci. Rep. 10, 12461. https://​doi.​org/​10.​1038/​s41598-​020-​69367-3 (2020).
2. Morone, G. et al. Rehabilitative devices for a top-down approach. Expert Rev. Med. Devices 16, 187–195 (2019).
3. Morone, G. et al. Robot-assisted gait training for stroke patients: Current state of the art and perspectives of robotics. Neuropsy-
chiatr. Dis. Treat 13, 1303–1311. https://​doi.​org/​10.​2147/​ndt.​S1141​02 (2017).
4. Cheng, P. Y. & Lai, P. Y. Comparison of Exoskeleton Robots and End-Effector Robots on Training Methods and Gait Biomechanics
258–266 (Springer, 2013).
5. Mehrholz, J. & Pohl, M. Electromechanical-assisted gait training after stroke: A systematic review comparing end-effector and
exoskeleton devices. J. Rehabil. Med. 44, 193–199 (2012).
6. Molteni, F. et al. Gait recovery with an overground powered exoskeleton: A randomized controlled trial on subacute stroke subjects.
Brain Sci. https://​doi.​org/​10.​3390/​brain​sci11​010104 (2021).
7. Karunakaran, K. K. et al. Effect of robotic exoskeleton gait training during acute stroke on functional ambulation. NeuroRehabilita-
tion 46, 493–503 (2021).
8. Turner, D. L., Ramos-Murguialday, A., Birbaumer, N., Hoffmann, U. & Luft, A. Neurophysiology of robot-mediated training and
therapy: A perspective for future use in clinical populations. Front. Neurol. 4, 184. https://d
​ oi.o
​ rg/1​ 0.3​ 389/f​ neur.2​ 013.0​ 0184 (2013).
9. Park, C. H., Hwang, J. S. & You, J. S. H. Comparative effectiveness of robot-interactive gait training with and without ankle robotic
control in patients with brain damage. J. Mech. Med. Biol. 1, 2140035 (2021).
10. Mooney, L. M. & Herr, H. M. Biomechanical walking mechanisms underlying the metabolic reduction caused by an autonomous
exoskeleton. J. Neuroeng. Rehabil. 13, 4. https://​doi.​org/​10.​1186/​s12984-​016-​0111-3 (2016).

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 11

Vol.:(0123456789)
 www.nature.com/scientificreports/

11. Calabrò, R. S. et al. Robotic gait rehabilitation and substitution devices in neurological disorders: where are we now?. Neurol. Sci.
37, 503–514. https://​doi.​org/​10.​1007/​s10072-​016-​2474-4 (2016).
12. Barroso, F. O. et al. Combining muscle synergies and biomechanical analysis to assess gait in stroke patients. J. Biomech. 63, 98–103.
https://​doi.​org/​10.​1016/j.​jbiom​ech.​2017.​08.​006 (2017).
13. Boudarham, J. et al. Variations in kinematics during clinical gait analysis in stroke patients. PLoS ONE 8, e66421. https://​doi.​org/​
10.​1371/​journ​al.​pone.​00664​21 (2013).
14. Li, S., Francisco, G. E. & Zhou, P. Post-stroke hemiplegic gait: New perspective and insights. Front. Physiol. 9, 1021. https://​doi.​
org/​10.​3389/​fphys.​2018.​01021 (2018).
15. Olney, S. J. & Richards, C. Hemiparetic gait following stroke. Part I: Characteristics. Gait Posture 4, 136–148. https://​doi.​org/​10.​
1016/​0966-​6362(96)​01063-6 (1996).
16. Yelnik, A., Albert, T., Bonan, I. & Laffont, I. A clinical guide to assess the role of lower limb extensor overactivity in hemiplegic
gait disorders. Stroke 30, 580–585. https://​doi.​org/​10.​1161/​01.​str.​30.3.​580 (1999).
17. von Schroeder, H. P., Coutts, R. D., Lyden, P. D., Billings, E. Jr. & Nickel, V. L. Gait parameters following stroke: A practical assess-
ment. J. Rehabil. Res. Dev. 32, 25–31 (1995).
18. Duval, K., Luttin, K. & Lam, T. Neuromuscular strategies in the paretic leg during curved walking in individuals post-stroke. J.
Neurophysiol. 106, 280–290. https://​doi.​org/​10.​1152/​jn.​00657.​2010 (2011).
19. Balaban, B. & Tok, F. Gait disturbances in patients with stroke. PM&R 6, 635–642 (2014).
20. Cruz-Montecinos, C. et al. Changes in muscle activity patterns and joint kinematics during gait in hemophilic arthropathy. Front.
Physiol. https://​doi.​org/​10.​3389/​fphys.​2019.​01575 (2020).
21. Fregly, B. J., Boninger, M. L. & Reinkensmeyer, D. J. Personalized neuromusculoskeletal modeling to improve treatment of mobil-
ity impairments: A perspective from european research sites. J. Neuroeng. Rehabil. 9, 18. https://​doi.​org/​10.​1186/​1743-​0003-9-​18
(2012).
22. Patten, C., Lexell, J. & Brown, H. E. Weakness and strength training in persons with poststroke hemiplegia: Rationale, method,
and efficacy. J. Rehabil. Res. Dev. 41, 293–312. https://​doi.​org/​10.​1682/​jrrd.​2004.​03.​0293 (2004).
23. Kim, C. M. & Eng, J. J. Magnitude and pattern of 3D kinematic and kinetic gait profiles in persons with stroke: Relationship to
walking speed. Gait Posture 20, 140–146 (2004).
24. van Kammen, K., Boonstra, A. M., van der Woude, L. H. V., Reinders-Messelink, H. A. & den Otter, R. Differences in muscle
activity and temporal step parameters between Lokomat guided walking and treadmill walking in post-stroke hemiparetic patients
and healthy walkers. J. Neuroeng. Rehabil. 14, 32. https://​doi.​org/​10.​1186/​s12984-​017-​0244-z (2017).
25. Lünenburger, L., Colombo, G. & Riener, R. Biofeedback for robotic gait rehabilitation. J. Neuroeng. Rehabil. 4, 1. https://​doi.​org/​
10.​1186/​1743-​0003-4-1 (2007).
26. Park, C. H. et al. Effects of innovative hip-knee-ankle interlimb coordinated robot training on ambulation, cardiopulmonary func-
tion, depression, and fall confidence in acute hemiplegia. NeuroRehabilitation 46, 577–587. https://​doi.​org/​10.​3233/​nre-​203086
(2020).
27. Goffredo, M. et al. Stroke gait rehabilitation: A comparison of end-effector, overground exoskeleton, and conventional gait training.
Appl. Sci. 9, 2627 (2019).
28. Molteni, F., Gasperini, G., Cannaviello, G. & Guanziroli, E. Exoskeleton and end-effector robots for upper and lower limbs reha-
bilitation: Narrative review. PM&R 10, S174–S188 (2018).
29. De Luca, A. et al. Recovery and compensation after robotic assisted gait training in chronic stroke survivors. Disabil. Rehabil.
Assist. Technol. 14, 826–838. https://​doi.​org/​10.​1080/​17483​107.​2018.​14669​26 (2019).
30. Park, I. J. et al. Comparative effects of different assistance force during robot-assisted gait training on locomotor functions in
patients with subacute stroke: An assessor-blind, randomized controlled trial. Am. J. Phys. Med. Rehabil. 98, 58–64. https://​doi.​
org/​10.​1097/​phm.​00000​00000​001027 (2019).
31. Park, J. H., Shin, Y. I., You, J. S. H. & Park, M. S. Comparative effects of robotic-assisted gait training combined with conventional
physical therapy on paretic hip joint stiffness and kinematics between subacute and chronic hemiparetic stroke. NeuroRehabilita-
tion 42, 181–190. https://​doi.​org/​10.​3233/​nre-​172234 (2018).
32. Kim, S. Y. et al. Effects of innovative walkbot robotic-assisted locomotor training on balance and gait recovery in hemiparetic
stroke: A prospective, randomized, experimenter blinded case control study with a four-week follow-up. IEEE Trans. Neural Syst.
Rehabil. Eng. 23, 636–642. https://​doi.​org/​10.​1109/​tnsre.​2015.​24049​36 (2015).
33. Yang, H. E. et al. Structural and functional improvements due to robot-assisted gait training in the stroke-injured brain. Neurosci.
Lett. 637, 114–119. https://​doi.​org/​10.​1016/j.​neulet.​2016.​11.​039 (2017).
34. Jung, J. H., Lee, N. G., You, J. H. & Lee, D. C. Validity and feasibility of intelligent Walkbot system. Electron. Lett. 45, 1016–1017
(2009).
35. Lerner, Z. F., Damiano, D. L. & Bulea, T. C. A lower-extremity exoskeleton improves knee extension in children with crouch gait
from cerebral palsy. Sci. Transl. Med. https://​doi.​org/​10.​1126/​scitr​anslm​ed.​aam91​45 (2017).
36. Alcobendas, M. M. et al. Lokomat robotic-assisted versus overground training within 3 to 6 months of incomplete spinal cord
lesion: Randomized controlled trial. Neurorehabil. Neural Repair. 26, 1058–1063. https://​doi.​org/​10.​1177/​15459​68312​448232
(2012).
37. Jin, L. H., Yang, S. S., Choi, J. Y. & Sohn, M. K. The effect of robot-assisted gait training on Locomotor function and functional
capability for daily activities in children with cerebral palsy: A single-blinded, randomized cross-over trial. Brain Sci. 10, 801
(2020).
38. Van, T. Q., Kim, S. H., Lee, K. H., Kang, S. C. & Ryu, J. H. in 2015 IEEE International Conference on Advanced Intelligent Mechatron-
ics (AIM). 652–657 (IEEE).
39. Duschau-Wicke, A., Caprez, A. & Riener, R. Patient-cooperative control increases active participation of individuals with SCI
during robot-aided gait training. J. Neuroeng. Rehabil. 7, 1–13 (2010).
40. Arene, N. & Hidler, J. Understanding motor impairment in the paretic lower limb after a stroke: A review of the literature. Top
Stroke Rehabil. 16, 346–356 (2009).
41. Winter, D. A. Biomechanics and Motor Control of Human Movement (Wiley, 2009).
42. Gregson, J. M. et al. Reliability of the tone assessment scale and the modified Ashworth scale as clinical tools for assessing poststroke
spasticity. Arch. Phys. Med. Rehabil. 80, 1013–1016 (1999).
43. Ghotbi, N., Ansari, N. N., Naghdi, S. & Hasson, S. Measurement of lower-limb muscle spasticity: Intrarater reliability of modified
modified Ashworth scale. J. Rehabil. Res. Dev. 48, 83–88 (2011).
44. Sullivan, K. J. et al. Fugl-Meyer assessment of sensorimotor function after stroke: Standardized training procedure for clinical
practice and clinical trials. Stroke 42, 427–432 (2011).
45. Bonnyaud, C. et al. Effect of a robotic restraint gait training versus robotic conventional gait training on gait parameters in stroke
patients. Exp. Brain Res. 232, 31–42. https://​doi.​org/​10.​1007/​s00221-​013-​3717-8 (2014).
46. Naro, A. et al. Breakthroughs in the spasticity management: Are non-pharmacological treatments the future?. J. Clin. Neurosci.
39, 16–27. https://​doi.​org/​10.​1016/j.​jocn.​2017.​02.​044 (2017).
47. Little, V. L., McGuirk, T. E. & Patten, C. Impaired limb shortening following stroke: What’s in a name?. PLoS ONE 9, e110140.
https://​doi.​org/​10.​1371/​journ​al.​pone.​01101​40 (2014).

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 12

Vol:.(1234567890)
www.nature.com/scientificreports/

48. Hyngstrom, A., Onushko, T., Chua, M. & Schmit, B. D. Abnormal volitional hip torque phasing and hip impairments in gait post
stroke. J. Neurophysiol. 103, 1557–1568. https://​doi.​org/​10.​1152/​jn.​00528.​2009 (2010).
49. Akbas, T. Delineating abnormal coordination patterns in post-stroke gait: a multidisciplinary approach Ph.D thesis, The University
of Texas at Austion, (2019).
50. Roy, A., Krebs, H. I., Barton, J. E., Macko, R. F. & Forrester, L. W. M., R. M. in 2013 IEEE International Conference on Robotics and
Automation. 2175–2182 (IEEE).
51. Swaminathan, K. & Krebs, H. I. in 2015 IEEE International Conference on Rehabilitation Robotics (ICORR). 555–558 (IEEE).
52. Kim, C. M. & Eng, J. J. The relationship of lower-extremity muscle torque to locomotor performance in people with stroke. Phys.
Ther. 83, 49–57 (2003).
53. Al-Chalabi, M. & Alsalman, I. Neuroanatomy, posterior column (dorsal column). (2018).
54. Niu, J. et al. Modality-based organization of ascending somatosensory axons in the direct dorsal column pathway. J. Neurosci. 33,
17691–17709 (2013).
55. Grillner, S. Biological pattern generation: The cellular and computational logic of networks in motion. Neuron 52, 751–766 (2006).
56. Jahn, K. et al. Supraspinal locomotor control in quadrupeds and humans. Prog. Brain Res. 171, 353–362 (2008).
57. Gassert, R. & Dietz, V. Rehabilitation robots for the treatment of sensorimotor deficits: A neurophysiological perspective. J. Neu-
roeng. Rehabil. 15, 1–15 (2018).
58. Brunnstrom, S. Motor testing procedures in hemiplegia: Based on sequential recovery stages. Phys. Ther. 46, 357–375 (1966).
59. Peng, Q. et al. Quantitative evaluations of ankle spasticity and stiffness in neurological disorders using manual spasticity evaluator.
J. Rehabil. Res. Dev. 48, 473 (2011).
60. You, S. H. et al. Virtual reality–induced cortical reorganization and associated locomotor recovery in chronic stroke: An experi-
menter-blind randomized study. Stroke 36, 1166–1171 (2005).

Acknowledgements
Please note that the part of the primary clinical outcomes in a randomised clinical trial (NCT03554642) were
reported in NeuroRehabilitation (PMID: 32538882). Kinematic and kinetic outcome data were published in the
present study.

Author contributions
C.P., J.S.H.Y., M.O.P., and D.E. were involved in the study design. M.O.P., D.E., K.F., and A.B. performed the
experiments and collected and analyzed them. C.P., J.S.H.Y. performed the analysis and wrote the manuscript.
C.P., J.S.H.Y., M.O.P. and D.E. participated in the document’s revision. All authors read and approved the final
manuscript.

Funding
This research received financial and administrative support from the Korea Institute for Robot Industry
Advancement Project (Grant No. ROBOT16-A003), Korea Health Industry Development Institute (Grant no.
HI18C1687000020) and the Brain Korea 21 PLUS Project (Grant No. 2019-51-0018) to the Department of Physi-
cal Therapy, Graduate School, Yonsei University. This manuscript has been submitted solely to this journal and
has not been published or submitted elsewhere.

Competing interests
The authors declare no competing interests.

Additional information
Correspondence and requests for materials should be addressed to J.H.Y.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.

© The Author(s) 2021

Scientific Reports | (2021) 11:22823 | https://doi.org/10.1038/s41598-021-01959-z 13

Vol.:(0123456789)